Threading using neural nEtwork (TUNE): the measure of protein sequence-structure compatibility

Area of Science:

• Computational Biology
• Structural Bioinformatics

Background:

• Fold recognition programs align protein sequences to 3D structures for prediction.
• Accurate sequence-structure alignment is crucial for determining protein 3D structure and biological function.

Purpose of the Study:

• To introduce a new threading scoring function for assessing protein sequence-structure compatibility.
• To utilize an artificial neural network (ANN) model for predicting amino acid side-chain compatibility within structural environments.

Main Methods:

• An artificial neural network model was trained to predict the compatibility of amino acid side-chains with their local structural environments.
• Log-odds scores derived from predicted probabilities were used to construct protein sequence-structure alignments.
• The model's performance was evaluated using a residue-level structural description.

Main Results:

• The developed scoring function demonstrates comparable performance to pseudo-energy functions utilizing atom-level descriptions.
• The model outperforms existing residue-level scoring functions in discriminating native from decoy protein structures.
• The ANN approach provides a robust method for sequence-structure compatibility scoring.

Conclusions:

• The novel ANN-based scoring function offers an effective approach for protein fold recognition.
• This method enhances the accuracy of predicting protein 3D structures and inferring biological functions.
• The C++ source code for the neural network model is publicly available for research use.